GPU memory

[dimvarsamis]

Hello and thanks for this very nice job.
I want to ask if the code could be refactored and use Distributed Data Parallel instead of Data Parallel. Or if you have a tip for me about the implementation, so i can do it.

thanks